Skip to content

Gandr TTS in any MCP agent.

One install gives Claude Code, Cursor, or any MCP-compatible agent the Gandr docs and a Gandr TTS client. Ask it to add Gandr to your project and it writes the integration for your stack.

Ask Claude CodeAsk ChatGPT

The whole integration

Install, then add the server to your MCP client.

Integration
pip install gandr-mcp

# Claude Code:
claude mcp add gandr -e GANDR_API_KEY=gnd_... -- gandr-mcp

# other MCP clients, add to the config:
{"mcpServers":{"gandr":{"command":"gandr-mcp","env":{"GANDR_API_KEY":"gnd_..."}}}}

Restart the client, then ask: "Add Gandr text to speech to this project." The key comes from your Gandr console.

What it needs

What it takes

One packagepip install gandr-mcp
A keyGANDR_API_KEY in the server env
An MCP clientClaude, Cursor, or anything that speaks MCP
Python3.9 or newer

Why an MCP agent needs speech

An agent that only returns text is half a tool: it can draft the reply but it cannot say it. With the Gandr MCP server, the agent finishes the job by speaking the answer in any of the 23 languages and one of the stock voices, or a cloned voice.

The server runs as its own process and holds the key, so the agent never sees a credential. Four tools speak: synthesize text, list voices, list languages, and check usage. Four more read the live Gandr docs and hand your coding agent working setup code for Python, Node, LiveKit, Pipecat, Vapi, LangChain, CrewAI, Haystack, the AI SDK, n8n, Dify, the OpenAI-compatible endpoint and raw REST.

What the flat stream changes

MCP agents get invoked a lot. A per-character meter makes every spoken answer a line item someone questions; the flat per-stream price means the agent can talk as much as the work needs and nothing is counted.

  • One stream carries every utterance, however many the agent makes
  • No per-character meter, so the agent is never told to say less
  • The key stays in the server env, off the agent conversation

Notes

Does the MCP server need its own process?

Yes. The server runs as a subprocess and holds the key in its environment, so your MCP client never handles a credential.

Which voices can the agent use?

Any of the stock voices by gandr- id, or a cloned voice from the console. The same 23 languages are available to every tool.

A key and one stream, and any agent in your stack can speak.

Get a key

Full API reference , gandr.ai/docs